SCEC Award Number 24123 View PDF
Proposal Category Collaborative Research Project (Multiple Investigators / Institutions)
Proposal Title Earthquake Ground Motion Simulation in San Francisco Bay Area via Heavy-tailed VAEs and Contrastive Learning
Investigator(s)
Name Organization
Michael Mahoney International Computer Science Institute Maxime Lacour University of California, Berkeley Rie Nakata International Computer Science Institute Nori Nakata Massachusetts Institute of Technology Pu Ren Lawrence Berkeley National Lab
SCEC Milestones C2-2, D3-3, C1,2,3-1, C1-2 SCEC Groups GM, RC, Seismology
Report Due Date 03/15/2025 Date Report Submitted 05/20/2025
Project Abstract
 Predicting high-fidelity ground motions for future earthquakes is crucial for seismic hazard assessment and infrastructure resilience. Conventional empirical simulations suffer from sparse sensor distribution and geographically localized earthquake locations, while physics-based methods are computationally intensive and require accurate representations of Earth structures and earthquake sources. We propose a novel artificial intelligence (AI) simulator, Conditional Generative Modeling for Ground Motion (CGM-GM), to synthesize high-frequency and spatially continuous earthquake ground motion waveforms. CGM-GM leverages earthquake magnitudes and geographic coordinates of earthquakes and sensors as inputs, learning complex wave physics and Earth heterogeneities, without explicit physics constraints. This is achieved through a probabilistic autoencoder that captures latent distributions in the time-frequency domain and variational sequential models for prior and posterior distributions. We evaluate the performance of CGM-GM using small-magnitude earthquake records from the San Francisco Bay Area, a region with high seismic risks. CGM-GM demonstrates a strong potential for outperforming a state-of-the-art non-ergodic empirical ground motion model and shows great promise in seismology and beyond.
Intellectual Merit This project directly supports SCEC’s goals by developing a machine learning-based generative model to enhance ground motion simulation capabilities essential for seismic hazard analysis. By capturing spatial heterogeneities and extending generalization across magnitudes, the model addresses SCEC priorities of reducing ground motion uncertainty and bridging observational gaps. Integration with the SCEC Broadband Platform ensures reproducibility, open access, and ensemble-based uncertainty quantification. This interdisciplinary research unites expertise in seismology and machine learning, advancing SCEC’s mission to synthesize data and simulation methods, and to train students and early-career researchers in innovative, ML-driven approaches to earthquake science.
Broader Impacts This project contributes to public safety and infrastructure resilience by improving access to realistic ground motion simulations through open-source tools and a curated dataset for the San Francisco Bay Area. By integrating advanced ML with seismological modeling, it enables broader scientific engagement across disciplines. The project trained one postdoctoral researcher in cross-disciplinary methods, equipping them with expertise at the interface of earthquake science and machine learning. Continued outreach and collaborative development will support early-career researchers and promote inclusive participation. This work strengthens the foundation for data-driven hazard assessment and accelerates innovation within the SCEC community and beyond.
Project Participants Rie Nakata (ICSI), Pu Ren (LBL), Nori Nakata (ICSI), Michael Mahoney (ICSI)
Exemplary Figure Figure 2: Illustrative examples of generated FAS maps.
(a)–(c) show 10 Hz Fourier Amplitude Spectra (FAS) maps from the non-ergodic GMM, CGM-baseline, and CGM-GM. The red star marks the epicenter (M3.84, depth 7.94 km; 37°51.61′N, 122°15.61′W), and the blue triangle indicates the observation station within the SFBA region. (d)–(e) display across 2–15 Hz for all recordings, comparing CGM-GM, CGM-baseline, and GMM. Solid lines denote mean residuals, and shaded areas represent ±1 standard deviation, highlighting model performance and variability across frequencies.
Linked Publications

Add missing publication or edit citation shown. Enter the SCEC project ID to link publication.